Fish individual identification plays an important role in the aquaculture industry. However, due to the complex underwater environment and the diversity of individual fish, the existing convolutional neural network is challenging to meet both high-precision and real-time identification of individual fish. Targeting the problem of low accuracy of individual fish recognition in complex underwater environment, this paper proposed a fusion and improved YoloV4 and FaceNet based on deep learning technology to identify underwater fish individuals. This model adds the CBAM module to the backbone of the Yolo V 4 algorithm. In the network section, this paper first uses the YoloV4 algorithm to extract and train the feature of the data set, and then uses the FaceNet algorithm to learn and predict the samples. The experimental results show that in the fish detection task, the proposed algorithm shows better performance than the original YoloV4 algorithm and effectively improves the accuracy of fish identification while ensuring real-time performance.